Prediction and optimization of abrasive wear loss of ultrahigh strength martensitic steel using response surface methodology, Harris Hawk and artificial neural network

Three-body abrasion wear problem in real industry application has been significantly reduced by replacing the heat-treatable steels with the newly developed ultrahigh strength martensitic steel. The wear performance under these conditions of one such steel, namely JFE EH400 was investigated in the present work. The input process parameters selected at different levels were employed to formulate the design matrix. Accordingly, 30 number of real time abrasion wear experiments were performed using dry sand rubber wheel test. The experimental results obtained were used to develop the quadratic model using Response Surface Methodology. Further, the prediction effectiveness was verified using Analysis of Variance. The results showed that the effect of load on wear loss was found to be most significant followed by the number of revolutions, flow rate and rotational speed. Moreover, for the validation of the performance obtained from statistical analysis, experimental data was employed to build the prediction models using Neural Networks (NNs). The proposed improved Generalised Regression-NN (GR-NN) was found to be an efficient and explorative predictive model in comparison to the Levenberg–Marquardt Perceptron-NN and Adaptive Linear-NN. The GR-NN was found to be most accurate owing to the minimum error as compared to other prediction models. The prediction ability of the GR-NN model with Harris Hawk Optimization was found to be better as compared to quadratic model, which was further validated using Scanning Electron Microscopy (SEM). The proposed model is efficient, accurate and encouraging for the prediction of wear loss in the industrial applications involving abrasion wear conditions.